Several techniques for handling missing observations in the statistical analysis of large data sets have been proposed. All of the existing techniques require certain assumptions about the underlying model; generally these include the assumption that the missing data are randomly generated and randomly dispersed within the data set. This project is assessing the robustness of the standard techniques to variations in the underlying assumptions, such as nonrandomness and high intercorrelations within the data set. The project will be accomplished using computer simulation on our HP-1000 minicomputer. No further work is expected on this project.